Direct imaging is likely the best way to characterize the atmospheres of Earth-sized exoplanets in the habitable zone of Sun-like stars. Previously, Stark et al. (2014, 2015, 2016) estimated the Earth twin yield of future direct imaging missions, such as LUVOIR and HabEx. We take an important next step by extending this analysis to other types of planets, which will act as false positives for Earth twins. We define an Earth twin as any exoplanet within half an e-folding of 1 AU in semi-major axis and 1 RE in planetary radius, orbiting a G dwarf. Using Monte Carlo analyses, we quantify the biases and planetary false positive rates of Earth searches. That is, given a pale dot at the correct projected separation and brightness to be a candidate Earth, what are the odds that it is, in fact, an Earth twin?Our notional telescope has a diameter of 10 m, an inner working angle of 3λ/D (62 mas at 1.0 micron), and an outer working angle of 10λ/D (206 mas). With no precursor knowledge and one visit per star, we detect many more un-Earths-77% of detected candidate Earths have an unEarthlike radius and/or semi-major axis, and their mean radius is 2.3 R_E, a sub-Neptune. The odds improve if we image every planet at its optimal orbital phase, helped either by precursor knowledge or multi-epoch direct imaging. 47% of detected Earth twin candidates are false positives in this targeted scenario, with a mean radius of 1.7 R_E. The false positive rate is robust to stellar spectral type and the assumption of circular orbits. The majority of false positives will be ""big and dark"" planets with large radii and low apparent albedos. Indeed, the radius-albedo degeneracy is the ultimate challenge in reflected light direct imaging. The false positive rate is <=50\% unless all planets have the same albedo and we know that value a priori. We might reduce the degeneracy via a mass-radius relation, if we know planetary mass from radial velocity or astrometry.